A methodology for stochastic analysis of share prices as Markov chains with finite states
- Felix Okoe Mettle^{1},
- Enoch Nii Boi Quaye^{1}Email author and
- Ravenhill Adjetey Laryea^{2}
https://doi.org/10.1186/2193-1801-3-657
© Mettle et al.; licensee Springer. 2014
Received: 19 August 2014
Accepted: 24 October 2014
Published: 6 November 2014
Abstract
Price volatilities make stock investments risky, leaving investors in critical position when uncertain decision is made. To improve investor evaluation confidence on exchange markets, while not using time series methodology, we specify equity price change as a stochastic process assumed to possess Markov dependency with respective state transition probabilities matrices following the identified state pace (i.e. decrease, stable or increase). We established that identified states communicate, and that the chains are aperiodic and ergodic thus possessing limiting distributions. We developed a methodology for determining expected mean return time for stock price increases and also establish criteria for improving investment decision based on highest transition probabilities, lowest mean return time and highest limiting distributions. We further developed an R algorithm for running the methodology introduced. The established methodology is applied to selected equities from Ghana Stock Exchange weekly trading data.
Keywords
Markov process Transition probability matrix Limiting distribution Expected mean return time Markov chainBackground
Stock market performance and operation has gained recognition as a significantly viable investment field within financial markets. We most likely find investors seeking to know the background and historical behavior of listed equities to assist investment decision making. Although stock trading is noted for its likelihood of yielding high returns, earnings of market players in part depend on the degree of equity price fluctuations and other market interactions. This makes earnings very volatile, being associated with very high risks and sometimes significant losses.
In stochastic analysis, the Markov chain specifies a system of transitions of an entity from one state to another. Identifying the transition as a random process, the Markov dependency theory emphasizes "memoryless property" i.e. the future state (next step or position) of any process strictly depends on its current state but not its past sequence of experiences noticed over time. Aguilera et al. (1999) noted that daily stock price records do not conform to usual requirements of constant variance assumption in conventional statistical time series. It is indeed noticeable that there may be unusual volatilities, which are unaccounted for due to the assumption of stationary variance in stock prices given past trends. To surmount this problem, models classes specified under the Autoregressive Conditional Heteroskedastic (ARCH) and its Generalized forms (GARCH) make provisions for smoothing unusual volatilities.
Against the characteristics of price fluctuations and randomness which challenges application of some statistical time series models to stock price forecasting, it is explicit that stock price changes over time can be viewed as a stochastic process. Aguilera et al. (1999) and Hassan and Nath (2005) respectively employed Functional Principal Component Analysis (FPCA) and Hidden Markov Model (HMM) to forecast stock price trend based on non-stationary nature of the stochastic processes which generate the same financial prices. Zhang and Zhang (2009) also developed a stochastic stock price forecasting model using Markov chains.
Varied studies (Xi et al.2012; Bulla et al.2010; Ammann and Verhofen2006; and Duffie and Singleton1993) have researched into the application of stochastic probability to portfolio allocation. Building on existing literature, we assume that stock price fluctuations exhibit Markov’s dependency and time-homogeneity and we specify a three state Markov process (i.e. price decrease, no change and price increase) and advance the methodology for determining the mean return time for equity price increases and their respective limiting distributions using the generated state-transition matrices. We further replicate the case for a two-state space i.e. decrease in price and increase in price. Based on the methodology, we hypothesize that;
Equity with the highest state transition probability and least mean return time will remain the best choice for an investor.
We explore model performance using weekly historical data from the Ghana Stock Exchange (GSE); we set up the respective transition probability matrix for selected stocks to test the model efficiency and use.
Review of theoretical framework
Definition of the Markov process
where t_{ n } < τ < t and S is the state space of the process {X (t)}.
A stochastic process with discrete state and parameter spaces which exhibits Markov dependency as in (3) is known as a Markov Process.
equations (2) and (4) are known as the Chapman-Kolmogorov equations for the process.
n-step transition probability matrix and n-step transition probabilities
If P is the transition probability matrix of a Markov chain {X_{ n }, n = 0, 1, 2, …} with state space S, then the elements of P^{ n } (P raised to the power n),${P}_{ij}^{\left(n\right)}\phantom{\rule{0.25em}{0ex}}i,j\u03f5S$ are the n-step transition probabilities where P_{ ij }^{(n)} is the probability that the process will be in state j at the n^{ th } step starting from state i.
which again can be seen as the (i, j)^{ th } element of the matrix product P^{ r }P = P^{r+1}. Hence by induction, P_{ ij }^{(n)} is the (i, j)^{ th } element of P^{ n }n = 2, 3, ….
To specify the model, the underlying assumption is stated about the identified n-step transition probability (stating without proof).
The transition probability matrix is accessible with existing state communication. Further, there exists recurrence and transience of states. States are also assumed to be irreducible and belong to one class with the same period which we take on the value 1. Thus the states are aperiodic.
Limiting distribution of a Markov chain
Where α = [α_{1}, α_{2}, …, α_{ m }] with 0 < α_{ j } < 1 and$\sum _{j=1}^{m}{\mathit{\alpha}}_{\mathit{j}}=1$. See Bhat (1984). The chain with this property is said to be ergodic and has a limiting distribution π. The transition probability matrix P of such a chain is primitive.
Recurrence and transience of state
and${\mu}_{ii}={\mu}_{i}={\displaystyle \sum _{t=1}^{\infty}t{f}_{ii}^{\left(t\right)}}$ is the mean recurrence time of state i if state i is recurrent.
A state i is said to be transient if and only if, starting from state i, there is a positive probability that the process may not eventually return to this state. This means f_{ ii }^{*} < 1
Model specification
Defining the problem (Equity price changes as a three-state Markov process)
Suppose the data in Additional file1 is uploaded as .csv, then R code for computing estimates in (12b) can be found in Additional file2 (three-state Markov Chain function column).
For a two-state Markov process
With the corresponding R algorithm shown in Additional file2 (two-state Markov Chain function column).
Generating eigen vectors for computation of limiting distributions
After the transition probabilities are obtained for both two-state and three-state chains, the R codes in the lower portions of columns one and two in Additional file2 were used to generate the respective eigen vectors for computation of limiting distributions.
Findings and discussions
Data structure and summary statistics
Data used for this paper are weekly trading price changes for five randomly selected equities on the Ghana Stock Exchange (GSE), each covering period starting from January 2012-December 2013. We obtain the weekly price changes using the relation d_{ t } = Y_{ t } - Y_{t-1} where Y_{ t } represents the equity closing price on week t and Y_{t-1} is the opening price for the immediate past week. The equities selected include Aluworks (ALW), Cal Bank (CAL), Ecobank Ghana (EBG), Ecobank Transnational Incorporated (ETI), and Fan Milk Ghana Limited (FML).
In all, 104 (52 weeks) observational data points where obtained. Summary statistics on all respective equities on the GSE are shown in Table 1. We present summaries on the respective number of weekly price decreases, no change in price and price increase. Descriptive statistics for each equity weekly price change is also shown.
Summary statistics on the weekly trading price change over the study period
Number of weekly price change | Weekly price change summary | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
Decrease | No change | Increase | Mean | SD | Max | Min | Skew. | Kurt. | Count | |
ALW | 15 | 77 | 12 | 0.00 | 0.01 | 0.01 | -0.04 | -2.30 | 14.23 | 104 |
AYRTN | 8 | 89 | 7 | 0.00 | 0.00 | 0.01 | -0.01 | -0.10 | 4.39 | 104 |
BOPP | 26 | 45 | 33 | 0.01 | 0.13 | 0.44 | -0.62 | -1.80 | 11.79 | 104 |
CAL | 27 | 40 | 37 | 0.00 | 0.03 | 0.12 | -0.07 | 1.69 | 7.26 | 104 |
EBG | 30 | 44 | 30 | 0.00 | 0.19 | 0.50 | -1.60 | -5.65 | 51.32 | 104 |
EGL | 21 | 46 | 37 | 0.01 | 0.06 | 0.39 | -0.25 | 1.05 | 15.79 | 104 |
ETI | 18 | 59 | 27 | 0.00 | 0.01 | 0.04 | -0.04 | -0.71 | 5.62 | 104 |
FML | 22 | 38 | 44 | 0.03 | 0.12 | 0.85 | -0.19 | 4.08 | 25.22 | 104 |
GCB | 25 | 37 | 42 | 0.02 | 0.13 | 0.79 | -0.41 | 1.68 | 12.51 | 104 |
GGBL | 5 | 51 | 48 | 0.04 | 0.10 | 0.73 | -0.20 | 3.84 | 22.70 | 104 |
GLD | 7 | 79 | 18 | 0.04 | 0.34 | 3.13 | -0.72 | 7.22 | 64.43 | 104 |
GOIL | 16 | 53 | 35 | 0.00 | 0.03 | 0.12 | -0.23 | -2.99 | 23.63 | 104 |
HFC | 8 | 75 | 21 | 0.01 | 0.03 | 0.27 | -0.08 | 5.76 | 47.80 | 104 |
MLC | 7 | 75 | 22 | 0.00 | 0.01 | 0.05 | -0.03 | 1.04 | 5.42 | 104 |
PBC | 13 | 81 | 10 | 0.00 | 0.01 | 0.04 | -0.02 | 1.53 | 11.02 | 104 |
PZC | 22 | 55 | 27 | 0.01 | 0.38 | 3.02 | -1.00 | 4.78 | 39.15 | 104 |
SCB | 38 | 40 | 26 | 0.14 | 1.30 | 9.54 | -4.19 | 4.45 | 30.38 | 104 |
SCBPREF | 11 | 87 | 6 | 0.00 | 0.01 | 0.01 | -0.03 | -2.35 | 9.67 | 104 |
SIC | 19 | 67 | 18 | 0.00 | 0.02 | 0.16 | -0.06 | 5.55 | 50.15 | 104 |
SOGEGH | 12 | 89 | 3 | 0.00 | 0.02 | 0.01 | -0.18 | -6.56 | 50.60 | 104 |
SWL | 9 | 84 | 11 | 0.01 | 0.45 | 3.15 | -2.00 | 2.28 | 27.27 | 104 |
TBL | 21 | 62 | 21 | 0.02 | 0.62 | 2.99 | -3.00 | 0.16 | 12.33 | 104 |
TLW | 16 | 56 | 32 | 0.23 | 0.97 | 6.56 | -1.97 | 3.77 | 19.38 | 104 |
TOTAL | 16 | 66 | 22 | 0.01 | 0.08 | 0.52 | -0.16 | 4.89 | 29.16 | 104 |
TRANSOL | 12 | 63 | 29 | 0.04 | 0.18 | 1.26 | -0.50 | 4.24 | 24.87 | 104 |
UNIL | 3 | 76 | 25 | 0.03 | 0.23 | 1.79 | -0.77 | 5.23 | 38.66 | 104 |
UTB | 12 | 87 | 5 | 0.00 | 0.01 | 0.02 | -0.02 | -1.30 | 6.18 | 104 |
Empirical results on model application (three-state Markov chain)
For the five randomly selected equities, the transition probabilities of the equities are presented as follows. These were obtained from equation (12a) defining${P}_{ij}=\frac{{n}_{ij}}{{n}_{i}}$ w.r.t. the three-state space Markov process. A 3 × 3 transition matrix is obtained for respective equities as defined by (12b).
Clearly,${\widehat{P}}_{ij}>0$ for all i, j = 0, 1, 2 indicating irreducibility of the chains for all equities. Hence state 0 for all the equities is aperiodic and since periodicity is a class property, the chains are aperiodic. These imply that the chains are ergodic and have limiting distributions.
From the plot, FML share is the best choice for the investor since the probability that it increases from a high price to another higher price is higher when compared to the other selected stocks. ALW recorded the least probability of transition within the period. Comparing CAL to EBG, the methodology shows that CAL shares maintain high probability of moving to higher prices as compared to EBG shares although the later started with high prices at inception.
Entries of the limiting distribution at for respective equities
Equity | Limiting distribution | ||
---|---|---|---|
α _{1} | α _{2} | α _{3} | |
ALW | 0.141509 | 0.745283 | 0.113208 |
CAL | 0.244980 | 0.406396 | 0.348625 |
EBG | 0.269568 | 0.443025 | 0.287407 |
ETI | 0.168470 | 0.586912 | 0.244618 |
FML | 0.198113 | 0.386792 | 0.415094 |
Empirical model application (the two-state Markov process)
Entries of two-state transition matrices for selected equities
Equities | P _{00} | P _{01} | P _{10} | P _{11} |
---|---|---|---|---|
1 - θ | θ | β | 1 - β | |
ALW | 0.133333 | 0.866667 | 0.142857 | 0.857143 |
CAL | 0.296296 | 0.703704 | 0.227848 | 0.772152 |
EBG | 0.433333 | 0.566667 | 0.210526 | 0.789474 |
ETI | 0.166667 | 0.833333 | 0.170455 | 0.829545 |
FML | 0.380952 | 0.619048 | 0.152941 | 0.847059 |
Expected mean return time for respective stocks
Equity | μ_{00} | μ_{11} |
---|---|---|
$\frac{\left(\mathit{\theta}\mathbf{+}\mathit{\beta}\right)}{\mathit{\beta}}$ | $\frac{\mathit{\theta}\mathbf{+}\mathit{\beta}}{\mathit{\theta}}$ | |
ALW | 1.1555556 | 7.4285714 |
CAL | 1.3837280 | 3.6060127 |
EBG | 1.5488889 | 2.8218623 |
ETI | 1.2009132 | 5.9772727 |
FML | 1.4497355 | 3.2235294 |
Conclusion
The Markov Process provides a credible approach for successfully analyzing and predicting time series data which reflect Markov dependency. The study finds that all states obtained communicate and are aperiodic and ergodic hence possessing limiting distributions. It is distinctive from Figures 1 and2 (expected return time and t-step state transition probabilities of equity price increases i.e. P_{ ij } transition from state 2 to state 2) that the investor gains good knowledge about the characteristics of the respective equities hence improving decision making in the light return maximization. With regards to the selected stocks, FML equity recorded the highest state transition probabilities, highest limiting distribution but the second lowest mean return time to price increases (i.e. 3.224 weeks).
Our suggested use of Markov chains as a tool for improving stock trading decisions indeed aids in improving investor knowledge and chances of higher returns given risk minimization through best choice decision. We showed that the proposed method of using Markov chains as a stochastic analysis method in equity price studies truly improves equity portfolio decisions with strong statistical foundation. In our future work, we shall explore the case of specifying an infinite state space for the Markov chains model in stock investment decision making.
Declarations
Authors’ Affiliations
References
- Aguilera MA, Ocaña AF, Valderrama JM Applied stochastic models in business and industry. In Stochastic modeling for evolution of stock prices by means of functional principal component analysis. John Wiley & Sons, Ltd., New York; 1999.Google Scholar
- Ammann M, Verhofen M Working paper series in Finance. No. 20. The effect of market regimes on style allocation 2006.http://www.finance.unisg.chGoogle Scholar
- Bhat UN Wiley series in Probability & Mathematical Statistics. Elements of applied stochastic processes 2nd edition. 1984.Google Scholar
- Bulla J, Mergner S, Bulla I, Sesboüé A, Chesneau C Munich Personal RePEc Archive. Markov-switching asset allocation: do profitable strategies exist? 2010. MPRA Paper no. 21154.http://mpra.ub.uni-muenchen.de/21154/Google Scholar
- Duffie D, Singleton JK: Simulated moments estimation of Markov models of asset prices. Econometrica 1993, 61(4):929-952. 10.2307/2951768View ArticleGoogle Scholar
- Hassan RM, Nath B Proceedings of the 2005 5th international conference on intelligent systems design and applications (ISDA’05). In Stock market forecasting using hidden Markov model: a new approach. IEEE Computer Society, Washington, DC, USA; 2005.Google Scholar
- Xi X, Mamon R, Davison M: A higher-order hidden Markov chain-modulated model for asset allocation. J Math Model Algorithm Oper Res 2012, 13(1):59-85.View ArticleGoogle Scholar
- Zhang D, Zhang X: Study on forecasting the stock market trend based on stochastic analysis method. Int J Bus Manage 2009, 4(6):163-170.Google Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.